Professional Documents
Culture Documents
To cite this article: Shuyuan Liu & Tat Leung Chan (2017) A coupled CFD-Monte Carlo method
for simulating complex aerosol dynamics in turbulent flows, Aerosol Science and Technology, 51:3,
269-281, DOI: 10.1080/02786826.2016.1260087
1. Introduction
flow field besides the complex aerosol dynamics that
Turbulent flows with complex aerosol dynamics of poly- particles are involved.
dispersed particles are encountered in many scientific Population balance equation (PBE) is used to describe
and engineering problems. Examples mainly include the the evolution of PSD of particles mathematically
exhaust particle formation and evolution in the wake of (Ramkrishna, 2000), which is a transport equation of
the studied ground vehicle (Chan et al. 2010), the particle number density function dependent on space
dynamics and dispersion of nanoparticles in urban coordinates, time, and particle size. Depending on actual
atmospheric environment (Yu et al. 2009; Kumar et al. aerosol dynamic processes, different source terms such
2011), the nanoparticle synthesis in turbulent reacting as coagulation, nucleation and growth may appear in
flows (Yu et al. 2008; Akroyd et al. 2011), and the for- PBE, as shown in Equation (4) of Section 2.1. Combining
mation of soot particles in combustion engines (Chan different source terms into different physical processes,
and Cheng 2007; Cenker et al. 2013; Pang et al. 2016). the PBE (i.e., in Equation (4)) is obviously an integro-dif-
Particles involved in the aerosol processes are character- ferential equation. There has been a large number of
ized by a polydispersed particle size distribution (PSD). research studies concerning the solution of PBE in
The control over PSD of the particles in turbulent flows spatially homogeneous domain i.e., zero dimensional
is of paramount significance in many industrial and PBE (Efendiev 2004; Yu et al. 2009; Chan et al. 2010;
engineering applications because PSD reveals the basic Zhou and Chan 2011; Geng et al. 2013; Yu and Chan
properties of product particles or emission particles, 2015; Liu et al. 2015).
which in turn determines the engineering application or However, in actual industrial and engineering applica-
environmental effect. The PSD is subject to turbulent tions, an inhomogeneous flow field is generally encountered
CONTACT Tat Leung Chan mmtlchan@polyu.edu.hk Department of Mechanical Engineering, The Hong Kong Polytechnic University, Kowloon,
Hong Kong.
Color versions of one or more of the figure in the article can be found online at www.tandfonline.com/uast.
© 2017 American Association for Aerosol Research
270 S. LIU AND T. L. CHAN
which has a profound impact on aerosol dynamic processes. required for the PBE. These PDF methods can be divided
Those processes are basically dependent on local flow field into three categories i.e. Eulerian particle method (Pope
variables (e.g., temperature, concentration). Thus the solu- 1981), Lagrangian particle method (Pope, 1985) and Euler-
tion for multidimensional PBE including convection and ian field method (Sabel’nikov and Soulard 2005). Both the
diffusion terms becomes significant for aerosol dynamics in advantages and disadvantages of the methods and possible
turbulent flows. Coupling the PBE of aerosol dynamics with improvements can easily be identified in the comparison
CFD method provides a very promising approach to deal between Eulerian and Lagrangian Monte Carlo PDF meth-
with the spatially inhomogeneous problems of aerosol ods (Mobus et al. 2001; Zhang and Chen 2007; Jaishree and
dynamics (Kruis et al. 2012; Zhao and Zheng 2013; Zhou Haworth 2012).
and Chan 2014; Zhou and He 2014; Akridis and Rigopoulos Lagrangian particle Monte Carlo algorithms (Pope
2015; Amokrane et al. 2016). In laminar flow, the coupling 1985) have been regarded as the mainstream approach for
of CFD to PBE can be easily accomplished via proper trans- solving PDF transport equations in most applications of
formation of PBE. However, in turbulent flows, the closure PDF methods to date (Jaishree and Haworth 2012). The
problems arise due to the effect of turbulence on aerosol PDF is represented by a great number of notional particles
dynamic processes (e.g., coagulation, nucleation and which evolve according to the prescribed stochastic differ-
growth) as such physical processes are highly dependent on ential equations (SDE), and weighted averages over the
the local field variables. Moreover, the relationship between particles in a small amount of neighboring grids are used
turbulence, particle properties and collision kernels of aero- to approximate the local mean quantities. As the mean
sol dynamics is not well understood due to the theoretical velocity and turbulence are required before advancing in
limitations and experimental difficulties (Reade and Collins every time step, it is necessary to couple the Lagrangian
2000; Balachandar and Eaton 2010). Thus particular atten- particle method with a conventional CFD solver to formu-
tion is paid to examine the effect of turbulence on aerosol late a hybrid Lagrangian particle/Eulerian mesh PDF
dynamics and the evolution of PSD of aerosol dynamics in method (Jaishree and Haworth 2012). The main advantage
the present study. of Lagrangian particle method relative to Eulerian PDF
Probability density function (PDF) methods based on a method is that the spatial-transport algorithm has much
PDF transport equation have been proposed and used to higher accuracy. The number of grid cells required for
deal with turbulence, scalar transport or combustion prob- equivalent accuracy is thus considerably smaller and the
lems in turbulent reactive flows for more than three decades total computational cost of Lagrangian PDF is only pro-
(Pope 1981, 1985; Valino 1998; Sabel’nikov and Soulard portional to the number of notional particles despite of the
2005; Meyer 2010; Pope and Tirunagari 2014; Consalvi and special care required for reducing statistical error. In the
Nmira 2016). Originated mainly from the work of Lundgren present study, a newly modified consistent hybrid Lagrang-
(1967), the PDF approach has then been improved and ian particle/Eulerian mesh PDF method based mainly on
extended by many eminent researchers. Monte Carlo meth- the work of Rigopoulos (2007) is presented for the coupled
ods for solving the joint scalar and velocity-scalar PDF were CFD-Monte Carlo simulation of aerosol dynamics in tur-
developed by Pope (1981, 1985) and were reviewed compre- bulent flows. The novelty of the present study is to investi-
hensively in the book by Fox (2003). The PBE-PDF trans- gate the effect of Rej number on the PSD of typical aerosol
port equation was firstly derived in Rigopoulos (2007), and dynamic processes in turbulent flows. The enhancing effect
the CFD-Lagrangian Monte Carlo method was also firstly of turbulence on aerosol dynamic processes as well as the
proposed and developed in Di Veroli and Rigopoulos (2010, competition and transition between different aerosol
2011), based on which the CFD-Lagrangian Monte Carlo dynamic processes are noted and analyzed. A new particle
method is further modified in the present study. Both PSD tracking method with high computational efficiency is
and particle number density distribution can then be treated adopted in the Monte Carlo simulation. Moreover, the
without additional assumptions for closure via the trans- operator splitting technique recently proposed by Liu and
ported PDF methods. The full PSD can thus be obtained Chan (2016) for solving simultaneous aerosol dynamic
directly with the transported PDF methods. The work of Di processes is also used to increase the computational effi-
Veroli and Rigopoulos (2010) was the first paper to develop ciency and accuracy.
a Lagrangian Monte Carlo method for the PDF of the PSD
for a reactive precipitation problem and this method was
used on an aerosol condensation problem in Di Veroli and 2. Methodology
Rigopoulos (2011). The concept was applied to LES in Pes-
2.1. Governing equations
mazoglou et al. (2014) and to deal with aggregation in Pes-
mazoglou et al. (2016). Moreover, complex and arbitrary The governing equations of the coupled fluid-particle
kernels of aerosol dynamics are allowed since no closure is dynamics in incompressible flows include the continuity
AEROSOL SCIENCE AND TECHNOLOGY 271
and momentum equations i.e. Navier–Stokes equa- contributes a source for particle with the size of v0
tions (1) and (2) as well as species transport equation i.e. which is the minimum particle size of the nuclei.
Eq. (3) (Rigopoulos, 2007). Coagulation term, which consists of two parts. The
first part is the birth part accounting for all the pos-
r u .x; t/ D 0 ½1 sible gains in particle number density with size of v
due to the coagulation between particle of size
@u .x; t/=@t C .u .x; t/ r /u .x; t/ (v-v0) and v0. The factor 1/2 is used to prevent dou-
½2
D n r 2 u .x; t/ ¡ r P .x; t/=r ble counting the coagulation events. The second
part is the death part accounting for all the loss of
@Ya =@t C r .u .x; t/Ya / particles with size of v due to all the possible combi-
½3
D Da r 2 Ya C v_ ðY1 ; Y2 ; . . . ; Ym Þ nations. K(u,v) is the coagulation kernel dependent
on the size of colliding pairs.
where u is the velocity of the carrier fluid phase, n is the
kinematic viscosity which is assumed constant, x is the 2.2. PDF transport equation formulation
coordinates of particles, t is the time, P is the pressure, r
The derivation of PDF formulation for aerosol dynamics
is the fluid density, Ya is the mass fraction of species (i.e.,
a D 1, 2,…, m), Da is the diffusion coefficient in compo- in turbulent flows is based on the work of Rigopoulos
(2007). First of all, a joint multipoint PDF of the mass
sition space and v_ is the source term determined by the
concentration and particle number density of species at
aerosol dynamic processes (i.e., coagulation, nucleation
full size range is introduced so that the expected PSD at
and growth) in the present study. An in-house time-
any point can be determined:
dependent Reynolds Averaged Navier Stokes (RANS)
code and k-e turbulence model are used together with Y.x; t/; n.v; x; t/ f .Y; N; x; t/ ½5
the following transported PDF method. More detailed
information about the in-house CFD code can be found where Y y1 ; y2 ; . . . ; ym and N n1 ; n2 ; . . . ; nn have
in Section S1 of the online supplementary information the following constraints:
(SI). ya < Ya .x; t/ < ya C dy
The PBE in terms of particle number density n(v, x, t), ½6
n < N.vi ; x; t/ < n C dn
a function of particle volume as well as of space coordi-
nates and time, can be written as Equation (4), in which and the normalization property is also satisfied:
n(v, x, t) is written as n(v) for simplicity, R1 R1
f .Y; N; x; t/dYdN D 1 ½7
@n.v/=@t C r .un.v// C @.G.Y1 ; Y2 ; . . . ; Ym ; v/n.v//=@v 0 0
where N is the set of particle number density, u is the veloc- process with mean <DWi > D 0 and covariance
ity of carrier fluid phase, G1 and G2 are the growth kernels, <DWi DWj > D Dtdi,j , CN is the characteristic scale con-
ai-j,j and ai,j are the coagulation kernels and v is the particle cerning micro mixing, W _ i ðN; YÞ is the source term for
volume. More details about the discretization of the con- accounting the variation of particle number density,
tinuous PBE can be found in Section S3 of the SI. G T is the turbulent diffusion coefficient, < U > is the
average velocity.
2.2.2. Final PDF transport equation It is noteworthy that a new particle tracking method
Based on full derivations in Rigopoulos (2007), the final is used in the simulation of coagulation events of par-
transport equation of PDF is expressed as follows: ticles. Instead of looping several neighboring grid cells
which contain a large amount of notional particles, the
@f Xm X m
@2 ð r Ya r Ya FÞ loop checking will be carried out within only one grid
¼ ¡ uf ¡ < u0 r F > ¡ D
@t a¼1 b¼1
@ya @yb cell. It implies that the loop checking will be per-
X n X n
@ ð r Ni r Ni FÞ
2
@ formed over all the particles located within the grid
Dp ½B1 ðy1 ; y2 ; . . . ; ym Þf cell to counter check if they are involved in coagula-
i¼1 j¼1
@n i @nj @y a
Figure 1. Three-dimensional schematic configuration of a cylindrical aerosol reactor (two-dimensional axisymmetric grid is generated in
the rectangular domain ABCD, not in scale).
3. Numerical simulation setup kinetic energy distributions in the aerosol reactor can be
found in Section S5 of the SI. Temperature is kept at
The configuration of the aerosol reactor in the present 300 K for all the computational domain. The standard
study is shown in Figure 1. The aerosol reactor is a cylin- k-e turbulence model is used for the turbulence compu-
drical aerosol reactor with radius of 0.225 m and length tation. After balancing the computational accuracy and
of 2 m. A nozzle with radius of 0.005 m and length of cost, 30 notional particles are allocated in each cell in the
0.01 m is located at the center of the cylindrical aerosol present Monte Carlo simulation. The simulation param-
reactor. Three partitions with height of 0.05 m and thick- eters for aerosol dynamics in turbulent flows are summa-
ness of 0.0005 m are evenly mounted in the axial direc- rized in Table 1. The computational grid of the present
tion inside the cylindrical aerosol reactor in order to coupled CFD-Monte Carlo method can be found in
enhance mixing. For aerosol dynamics in turbulent flows, Figure S1 of the SI.
particles with a volume fraction of 0.1 in the carrier gas
(air) is injected from the nozzle while gas phase (air) is
injected through the gas phase inlet. The injected par- 4. Results and discussion
ticles are potassium chloride (KCl) particles with a size 4.1. Coagulation in laminar and turbulent flows
range of 0.1 mm to 6.4 mm and density of 1980 kg/m3
(Calvo et al. 2013). The injection velocity of particulate As mentioned above in Section 3, the inlet of fluid and
phase ranges from 10 m/s to 40 m/s, the inlet velocity of particulate phases is switched to simulate laminar flow,
gas phase is 0.5 m/s. Coagulation kernel is obtained using i.e., the particles are injected into the aerosol reactor
the turbulent coagulation model by Saffman and Turner through the air inlet while air is injected through the
(1956). The nucleation rate is kept constant at 7.5 £ 106/
m3¢s with constant growth rate of 6.2 £ 10¡11 m/s. Table 1. Simulation parameters for aerosol dynamics in turbulent
flows.
Aerosol dynamics in laminar flow is also investigated in
the same aerosol reactor for comparison with aerosol Particulate phase Potassium chloride (KCl)
dynamics in turbulent flows, which is carried out by Particle density (kg/m3) 1980
switching the inlet conditions of velocity and species Initial particle size range (mm) 0.1¡6.4
Injection velocity of the jet (m/s) 10¡40
between the inlets of the nozzle and the gas phase. Corresponding Rej of the jet 3200¡12,800
Specifically, for the study of aerosol dynamics in the Velocity of continuous phase (m/s) 0.5
laminar flow, particles enter the gas phase inlet instead Coagulation model Turbulent kernel (Saffman and
Turner 1956)
of the nozzle as shown in Figure 1 while the nozzle is Nucleation rate (#/m3s) 7.5 £ 106
used for the entry of air at a low velocity of 0.5 m/s. Growth rate (m/s) 6.2 £ 10¡11
The number of cells in the 8112
Four injection velocities of particulate phase, i.e., computational domain
10 m/s, 15 m/s, 20 m/s, and 40 m/s are used to The number of simulation particles 30
per cell
investigate the effect of jet Reynolds (Rej) number on the Turbulence model k-e model
aerosol dynamics in turbulent flows. The turbulent
274 S. LIU AND T. L. CHAN
Figure 5. PSD in turbulent coagulation: Case A, Rej D 3200; Case Figure 6. Average diameter of particles for different aerosol
B, Rej D 4800; Case C, Rej D 6400; Case D, Rej D 12800 (the PBSM dynamic processes in turbulent flows.
results are obtained on the basis of the method proposed by
Hounslow et al. 1988).
particles in the aerosol reactor. The Rej number of 0 cor-
of the jet increasing from 3200 to 12800, the PSDs responds to the initial PSD of aerosol dynamics before
become wider and higher, with the peaks moving entering the aerosol reactor. As higher Reynold number
towards the higher end of particle size range, which of the jet is related to higher turbulent intensity in the
yields a more evenly distributed population of particles. present study, the increase of average diameter with Rej
By controlling proper turbulent flow field, the control number indicates that coagulation is significantly
over PSD of aerosol particles can be achieved, which is of enhanced, which may be explained by the enhanced mix-
importance in industrial and engineering applications. ing and more frequent collisions between particles.
Comparing the results obtained with the present modi- The particle number density (#/m3) contour in turbu-
fied LMC-PDF method and PBSM (Hounslow et al. lent coagulation under different Rej number is shown in
1988), an excellent agreement can be observed, which Figure 7. It can be seen that despite of the similar distri-
validates the reliability and computational efficiency of bution pattern of particle number density in the reactor,
this modified LMC-PDF method. The wider PSD with the particle number density decreased with increasing
increasing Rej number is also in accordance with the pre- Rej number due to coagulation, which suggests an
vious research studies by Reade and Collins (2000) and increase in average coagulation rate with the increase of
Garrick (2015), which reveals the effect of turbulence on
PSD. By increasing Rej numbers, turbulence induced
mixing is greatly enhanced, which leads to an enhanced
coagulation process. Meanwhile, the peaks of PSD mov-
ing towards the higher end of particle size spectrum with
increasing Rej number implies that more and larger par-
ticles are produced by enhanced coagulation. In other
words, as coagulation is enhanced, larger particles have a
better chance to appear due to more frequent collisions.
According to the definition of particle number based
average diameter of particles (Friedlander 2000), the
number averaged particle diameter can be obtained by
integrating the PSD over the entire particle diameter
range and then averaging over the total particle number
density of particles. Using this method, the initial average
diameter and average diameters of particles under
the turbulent coagulation with different Rej numbers are
obtained and shown in Figure 6. An increase in average
diameter from 0.124 mm to 0.245 mm can be observed
with increasing Re number from 0 to 12,800, which is an Figure 7. Particle number density (#/m3) contour in turbulent
obvious increase considering the large number of aerosol coagulation.
276 S. LIU AND T. L. CHAN
According to Garrick (2015), nucleation is the dominant from 4800 to 12,800. Comparing Figure 10 with Figure 6,
process until the collapse of the jet potential core and it can be found that the variation particle number density
turbulent mixing begins in the proximal region of the jet, contour is consistent with the variation of average parti-
where coagulation becomes the dominant process and cle diameter, which reflects the fact that coagulation
leads to a broader PSD. Although increasing Rej number becomes dominant with Rej number increasing from
may lead to stronger turbulence in order to enhance 3200 to 4800. However, as Rej number increases from
coagulation, high velocity of jet also causes the delay of 4800 to 12,800, nucleation becomes dominant in the
turbulent mixing in which nucleation becomes dominant competition with coagulation as the average diameter
in the proximal region of the jet. The reason accounting decreases continuously.
for the change of PSD mentioned above can be verified Figure 11 shows the variation of normalized particle
by the evolution of average diameter of particles with number density with axial distance. It can be seen
increasing Rej number as shown in Figure 6. It can be clearly that the normalized particle number density
seen that the average diameter of particles increases experiences significant change with Rej number
slightly with increasing Rej number from 3200 to 4800, increasing from 3200 to 12,800. The normalized parti-
and reaches the peak of around 0.8 mm on the PSD. This cle number density increases rapidly with Rej number
is due to the enhanced coagulation as nucleation leads to of 3200, which implies that nucleation is the dominant
smaller average diameter of particles. Moreover, Figure 6 mechanism resulting in a large number of new par-
shows that average diameter of particles decreases when ticles. As Rej number increases from 3200 to 4800, the
Rej number is larger than 4800, which yields a maximum normalized particle number density decreases signifi-
average diameter of particles at RejD 4800 for these four cantly, indicating that the dominant mechanism in the
studied cases. The Rej number at zero corresponds to the aerosol reactor changes from nucleation to coagulation
initial average diameter of particles before entering into as the latter reduces the total number of particles,
the aerosol reactor. which is also in accordance with the increase of aver-
The contour of particle number density for simulta- age particle diameter as shown in Figure 6. However,
neous coagulation and nucleation is shown in Figure 10. as the Rej number further increases from 4800 to
With Rej number increasing from 3200 to 4800, the 12,800, the normalized particle number density also
maxima of the particle number density decreases, imply- increases when compared with Rej number of 4800 at
ing that coagulation process which reduces particle den- the same axial position. These results indicate that
sity is dominant. An increase of the maxima of particle coagulation is weakened while nucleation is enhanced
number density is observed when Rej number increases when Rej number increases from 4800 to 12,800,
which is consistent with the findings in Garrick (2015)
that too high injection velocity causes the delay of tur-
bulent mixing and in turn reduces coagulation rate as
mentioned above.
Figure 10. Particle number density (#/m3) contour for simulta- Figure 11. Normalized particle number density for simultaneous
neous coagulation and nucleation in turbulent flows. coagulation and nucleation in turbulent flows.
278 S. LIU AND T. L. CHAN
4.4. Coagulation, nucleation and growth in 12800, implying that nucleation process is further
turbulent flows enhanced to be the dominant process.
The average diameters of particles for different Rej
Complex aerosol dynamics in turbulent flows has been
numbers are calculated in order to reveal the physical
widely encountered and is a challenging problem in the
mechanism behind the change of the PSD. Figure 6
numerical research. The PSDs of aerosol particles
shows an increase in average diameter for increasing Rej
experiencing simultaneous coagulation, nucleation and
number from 3200 to 4800. Whereas for Rej number
growth processes under different Rej numbers are
larger than 4800 (within these four studied Rej numbers),
shown in Figure 12. The results are also validated with
the average diameter of particles decreases with increas-
PBSM (Hounslow et al. 1988). Comparing with the
ing Re number. Based on the average diameter of par-
results of PSDs in Figure 9, it can be found that the
ticles calculated from the obtained PSD, it can be
shape of the PSDs is basically the same except for the
concluded that, with increasing Rej number from 4800 to
positions and values of some peaks on them. The PSD
6400, the change of the peak at around 0.6 mm is due to
at Rej D 3200 is also characterized by its peaks, which
pure nucleation as neither growth nor coagulation leads
are closer to the lower end of the particle size range
to the decrease in average diameter. Comparing the case
when compared to that in Figure 9. The different posi-
of turbulent coagulation and nucleation with this case in,
tions of peaks of the PSD should be due to the growth
it can be found that the particle number density of small
process since the other conditions in this case are the
particles increases with Rej number in both of the two
same with that in turbulent coagulation and nucleation
cases, which implies that nucleation is the dominant pro-
process except that growth process is introduced. With
cess among the simultaneous processes in the present
Rej number increasing from 3200 to 4800, a new peak
study.
at around 1.5 mm appears on the new PSD. This is
Figure 13 shows the particle number density contour
caused by the growth and coagulation processes as
for simultaneous coagulation, nucleation and growth in
nucleation process forms much smaller particles rather
turbulent flows. Consistent with the variation of average
than particles with diameter of 1.5 mm, which is shown
diameter of particles, the maxima of the particle number
by the increase of average particle diameter with Rej
density contour also increases with increasing Rej num-
number increasing from 3200 to 4800 in Figure 6. How-
ber from 3200 to 4800 and then decreases with Rej num-
ever, as Rej number increases from 4800 to 6400, the
ber for larger than 4800. The variation of particle
value of the peak at 1.5 mm decreases very slightly while
number density contour also verifies the effect of Re
the value of the peak at around 0.6 mm increases signif-
number on the competition between coagulation and
icantly, which may be due to the enhanced nucleation
nucleation. The dominant process changes from
process as nucleation produces more particles with the
smallest size. The value of the peak at 0.6 mm increased
sharply again with Rej number increasing from 6400 to
coagulation to nucleation with increasing Rej number high injection velocity (Garrick 2015). As for the case
from 3200 to 12,800 in the present study. Comparing with Rej number of 12,800, the low particle number den-
with the results in Figure 10, regions with high particle sity near the outlet of the aerosol reactor may be caused
number density appear within the core of the jet in by the dead zone between the second and the third parti-
Figures 13a–c, which may be due to the growth processes tions of the aerosol reactor where those particles are
taking place in these regions. The high particle number trapped.
density regions in Figure 13d, however, move from the More detailed information about computational accu-
core of the jet/nozzle injection to the zone between the racy and efficiency of the present modified CFD-MC
second and the third partitions, and close to the wall of method in Section S4 of the SI.
the aerosol reactor. This may be because this zone
becomes a dead zone of flow and mixing for
Rej D 12,800, which leads to weakened coagulation pro- 5. Conclusions
cess in this dead zone of flow and mixing. Typical simultaneous aerosol dynamic processes (i.e.,
Figure 14 presents the variation of normalized particle coagulation, nucleation and growth) are widely encoun-
number density with the axial distance. It can be seen tered in turbulent flows. They are investigated with the
that the variation of normalized particle number density present modified Lagrangian PDF approach based CFD-
is basically the same for all the four studied Rej numbers Monte Carlo method. The effect of Rej number on the
despite small differences, which increases first until a interactions between turbulence and aerosol dynamics is
peak appears between the second and the third partitions fully studied. The results reveal the significant impact of
of the aerosol reactor. Except for the case with Rej num- Rej number on both single aerosol process (e.g., coagula-
ber of 12,800, the normalized particle number density tion) and simultaneous competitive aerosol dynamic
after the peak decreases, and then basically remains sta- processes in turbulent flows. This newly modified CFD-
ble with slight increase at the outlet of the aerosol reactor Monte Carlo/PDF method renders an efficient method
for all the other three studied cases. The variation of nor- to deal with the interactions between turbulence and
malized particle number density for Rej number ranging aerosol dynamics. The full PSD of aerosol particles is
from 3200 to 6400 can be explained by the competition readily obtained. The enhancing effect of turbulence on
of multiple aerosol dynamic processes including coagula- coagulation is demonstrated by comparison of laminar
tion, nucleation and growth. However, the particle num- coagulation and turbulent coagulation. Further study of
ber density at the peak is so high that coagulation the effect of turbulence on the PSDs of simultaneous
process takes place, which reduces the particle number aerosol dynamics reveals the competition between aero-
density to a basically dynamic equilibrium between the sol dynamic processes, which also has significant impact
increasing factor (nucleation) and decreasing factor on the final PSD of particles. For simultaneous complex
(coagulation) of particle number density. Although coag- aerosol dynamics in turbulent flows, turbulence has an
ulation is not the dominant process in the four studied effect of broadening the PSD.
cases, it is shown in Figure 14 that once the Rej number
is larger than 4800, the normalized particle number den-
sity increases due to the delay of turbulent mixing at Funding
This work was supported by grants from the research student-
ship, Central Research Grant (Project No. G-YBF5) and
Mechanical Engineering Department of The Hong Kong Poly-
technic University. Last but not least, the financial support
from the National Natural Science Foundation of China (Proj-
ect No. 11572274) was also greatly appreciated for allowing the
authors’ further development and extension of this coupled
CFD-Monte Carlo method in the study of aerosol dynamics.
References
Akridis, P., and Rigopoulos, S. (2015). Modelling of Soot For-
mation in a Laminar Coflow Non-Premixed Flame with a
Detailed CFD-Population Balance Model. Procedia Eng.,
102:1274–1283.
Figure 14. Normalized particle number density for simultaneous Akroyd, J., Smith, A. J., Shirley, R., McGlashan, L. R., and
coagulation, nucleation and growth in turbulent flows. Kraft, M. (2011). A Coupled CFD-Population Balance
280 S. LIU AND T. L. CHAN
Approach for Nanoparticle Synthesis in Turbulent Reacting Walk Method and a Constant-Number Monte-Carlo
Flows. Chem. Eng. Sci., 66:3792–3805. Method for Aerosol Dynamics. Chem. Eng. Sci., 70:109–120.
Amokrane, A., Maass, S., Lamadie, F., Puel, F., and Charton, S. Kumar, P., Ketzel, M., Vardoulakis, S., Pirjola, L., and Britter,
(2016). On Droplets Size Distribution in a Pulsed Column. R. (2011). Dynamics and Dispersion Modelling of
Part I: In-Situ Measurements and Corresponding CFD-PBE Nanoparticles from Road Traffic in the Urban Atmospheric
Simulations. Chem. Eng. J., 296:366–376. Environment—A Review. J. Aerosol Sci., 42:580–603.
Balachandar, S., and Eaton, J. K. (2010). Turbulent Dispersed Liu, S. Y., Chan, T. L., and Zhou, K. (2015). A New Stochasti-
Multiphase Flow. Annu. Rev. Fluid Mech., 42:111–133. cally Weighted Operator Splitting Monte Carlo Method for
Calvo, A. I., Alves, C., Castro, A., Pont, V., Vicente, A. M., and Particle-Fluid Systems. ASME-ATI-UIT 2015 Conference
Fraile, R. (2013). Research on Aerosol Sources and Chemi- on Thermal Energy Systems: Production, Storage, Utiliza-
cal Composition: Past, Current and Emerging Issues. tion and the Environment (In Session: Computational
Atmos. Res., 120:1–28. Thermal-fluid Dynamics), May 17–20, Naples, Italy.
Cenker, E., Bruneaux, G., Pickett, L., and Schulz, C. (2013). Liu, S. Y., and Chan, T. L. (2016). A Stochastically Weighted
Study of Soot Formation and Oxidation in the Engine Operator Splitting Monte Carlo (SWOSMC) Method for
Combustion Network (ECN), Spray A: Effects of Ambient the Numerical Simulation of Complex Aerosol Dynamic
Temperature and Oxygen Concentration. SAE Int. J. Processes. Int. J. Numer. Methods Heat Fluid Flow,
Engines, 6:352–365. Accepted for publication. doi:10.1108/HFF-08-2015-0335
Chan, T. L., and Cheng, X. B. (2007). Numerical Modeling and Lundgren, T. S. (1967). Distribution Functions in the Statistical
Experimental Study of Combustion and Soot Formation in a Theory of Turbulence. Phys. Fluids, 10:969–975.
Direct Injection Diesel Engine. Energy Fuels, 21:1483–1492. Meyer, D. W. (2010). A New Particle Interaction Mixing
Chan, T. L., Liu, Y. H., and Chan, C.K. (2010). Direct Quadra- Model for Turbulent Dispersion and Turbulent Reactive
ture Method of Moments for the Exhaust Particle Forma- Flows. Phys. Fluids, 22:035103.
tion and Evolution in the Wake of the Studied Ground Mobus, H., Gerlinger, P., and Br€ uggemann, D. (2001). Com-
Vehicle. J. Aerosol Sci., 41:553–568. parison of Eulerian and Lagrangian Monte Carlo PDF
Consalvi, J. L., and Nmira, F. (2016). Transported Scalar PDF Methods for Turbulent Diffusion Flames. Combust. Flame,
Modeling of Oxygen-Enriched Turbulent Jet Diffusion 124:519–534.
Flames: Soot Production and Radiative Heat Transfer. Fuel, Pang, K. M., Karvounis, N., Walther, J. H., and Schramm, J.
178:37–48. (2016). Numerical Investigation of Soot Formation and
Di Veroli, G., and Rigopoulos, S. (2010). Modeling of Oxidation Processes Under Large Two-Stroke Marine Die-
Turbulent Precipitation: A Transported Population sel Engine-Like Conditions Using Integrated CFD-Chemi-
Balance-PDF Method. AIChE J., 56:878–892. cal Kinetics. Appl. Energ., 169:874–887.
Di Veroli, G., and Rigopoulos, S. (2011). Modeling of Aerosol Pesmazoglou, I., Kempf, A. M., and Navarro-Martinez, S.
Formation in a Turbulent Jet with the Transported Popula- (2014). Aerosol Nucleation in a Turbulent Jet Using Large
tion Balance Equation-Probability Density Function Eddy Simulations. Chem. Eng. Sci., 116:383–397.
Approach. Phys. Fluids, 23:043305. Pesmazoglou, I., Kempf, A. M., and Navarro-Martinez, S.
Efendiev, Y. (2004). Modelling and Simulation of Multi- (2016). Stochastic Modelling of Particle Aggregation. Int. J.
Component Aerosol Dynamics. Comput. Appl. Math., 23:401– Multiphase Flow, 80:118–130.
423. Pope, S. B. (1981). A Monte Carlo Method for PDF Equations
Fox, R. O. (2003). Computational Models for Turbulent React- of Turbulent Reactive Flow. Combust. Sci. Tech., 25:
ing Flows. Cambridge University Press, Cambridge. 159–174.
Friedlander, S. (2000). Smoke, Dust and Haze: Fundamentals of Pope, S. B. (1985). PDF methods for turbulent reactive flows.
Aerosol Dynamics. Second ed. Oxford University Press, Prog. Energy Combust. Sci., 11:119–192.
Oxford. Pope, S. B., and Tirunagari, R. (2014). Advances in Probability
Garrick, S. C. (2015). Growth mechanisms of nanostructured Density Function Methods for Turbulent Reactive Flows. In
Titania in turbulent reacting flows. J. Nanotechnology, 2015, Proceedings of the Nineteenth Australasian Fluid Mechan-
Article ID 642014:1–10. ics Conference, RMIT University, Melbourne.
Geng, J., Park, H., and Sajo, E. (2013). Simulation of Aerosol Ramkrishna, D. (2000). Population Balances: Theory and
Coagulation and Deposition Under Multiple Flow Regimes Applications to Particulate Systems in Engineering.
with Arbitrary Computational Precision. Aerosol Sci. Tech., Academic Press, San Diego.
47:530–542. Reade, W. C., and Collins, L. R. (2000). A Numerical Study of
Hounslow, M. J., Ryall, R. L., and Marshall, V. R. (1988). A the Particle Size Distribution of an Aerosol Undergoing
Discretized Population Balance for Nucleation, Growth, Turbulent Coagulation. J. Fluid Mech., 415:45–64.
and Aggregation. AIChE J., 34:1821–1832. Rigopoulos, S. (2007). PDF Method for Population Balance in
Jaishree, J., and Haworth, D. C. (2012). Comparisons of Turbulent Reactive Flow. Chem. Eng. Sci., 62:6865–6878.
Lagrangian and Eulerian PDF Methods in Simulations of Rigopoulos, S., and Jones, A. G. (2003). Finite-Element Scheme
Non-Premixed Turbulent Jet Flames with Moderate-to- for Solution of the Dynamic Population Balance Equation.
Strong Turbulence-Chemistry Interactions. Combust. The- AIChE J., 49:1127–1139.
ory Modell., 16:435–463. Sabel’nikov, V., and Soulard, O. (2005). Rapidly Decorrelating
Kruis, F. E., Wei, J. M., van der Zwaag, T., and Haep, S. (2012). Velocity-Field Model as a Tool for Solving one-Point
Computational Fluid Dynamics Based Stochastic Aerosol Fokker-Planck Equations for Probability Density Functions
Modelling: Combination of a Cell-Based Weighted Random of Turbulent Reactive Scalars. Phys. Rev. E, 72:016301.
AEROSOL SCIENCE AND TECHNOLOGY 281
Saffman, P. G., and Turner, J. S. (1956). On the Collision of Zhang, Z., and Chen, Q. (2007). Comparison of the Eulerian
Drops in Turbulent Clouds. J. Fluid Mech., 1:16–30. and Lagrangian Methods for Predicting Particle Transport
Valino, L. (1998). Field Monte Carlo Formulation for Calculat- in Enclosed Spaces. Atmos. Environ., 41:5236–5248.
ing the Probability Density Function of a Single Scalar in a Zhao, H. B., and Zheng, C. G. (2013). A Population Balance-
Turbulent Flow. Flow Turbulence Combust., 60:157–172. Monte Carlo Method for Particle Coagulation in Spatially
Yu, M. Z., Lin, J. Z., and Chan, T. L. (2008). Effect of Inhomogeneous Systems. Comput. Fluids, 71:196–207.
Precursor Loading on Non-Spherical TiO2 Nanoparticle Zhou, K., and Chan, T. L. (2011). Simulation of Homogeneous
Synthesis in a Diffusion Flame Reactor. Chem. Eng. Sci., Particle Nucleation in a Free Turbulent Jet. Aerosol Sci.
63:2317–2329. Technol., 45:973–987.
Yu, M. Z., Lin, J. Z., and Chan, T. L. (2009). Numerical Simula- Zhou, K., and Chan, T. L. (2014). Analytical Approximation
tion for Nucleated Vehicle Exhaust Particulate Matters via Schemes for Mean Nucleation Rate in Turbulent Flows.
the TEMOM/LES Method. Int. J. Mod. Phys. C, 20:399–421. Aerosol Sci. Technol., 48:459–466.
Yu, M. Z., and Chan, T. L. (2015). A Bimodal Moment Method Zhou, K., and He, Z. (2014). Monte Carlo Simulation of
Model for Submicron Fractal-Like Agglomerates Undergo- Aerosol Evolution in a Planar Mixing Layer. Int. J. Numer.
ing Brownian Coagulation. J. Aerosol Sci., 88:19–34. Methods Heat Fluid Flow, 24:1769–1781.